Pull ceph updates from Ilya Dryomov:
"A big pile of assorted fixes and improvements for the filesystem with
nothing in particular standing out, except perhaps that the fact that
the MDS never really maintained atime was made official and thus it's
no longer updated on the client either.
We also have a MAINTAINERS update: Jeff is transitioning his
filesystem maintainership duties to Xiubo"
* tag 'ceph-for-5.19-rc1' of https://github.com/ceph/ceph-client: (23 commits)
MAINTAINERS: move myself from ceph "Maintainer" to "Reviewer"
ceph: fix decoding of client session messages flags
ceph: switch TASK_INTERRUPTIBLE to TASK_KILLABLE
ceph: remove redundant variable ino
ceph: try to queue a writeback if revoking fails
ceph: fix statfs for subdir mounts
ceph: fix possible deadlock when holding Fwb to get inline_data
ceph: redirty the page for writepage on failure
ceph: try to choose the auth MDS if possible for getattr
ceph: disable updating the atime since cephfs won't maintain it
ceph: flush the mdlog for filesystem sync
ceph: rename unsafe_request_wait()
libceph: use swap() macro instead of taking tmp variable
ceph: fix statx AT_STATX_DONT_SYNC vs AT_STATX_FORCE_SYNC check
ceph: no need to invalidate the fscache twice
ceph: replace usage of found with dedicated list iterator variable
ceph: use dedicated list iterator variable
ceph: update the dlease for the hashed dentry when removing
ceph: stop retrying the request when exceeding 256 times
ceph: stop forwarding the request when exceeding 256 times
...
1, mount with wsync.
2, create a file with O_RDWR, and the request was sent to mds.0:
ceph_atomic_open()-->
ceph_mdsc_do_request(openc)
finish_open(file, dentry, ceph_open)-->
ceph_open()-->
ceph_init_file()-->
ceph_init_file_info()-->
ceph_uninline_data()-->
{
...
if (inline_version == 1 || /* initial version, no data */
inline_version == CEPH_INLINE_NONE)
goto out_unlock;
...
}
The inline_version will be 1, which is the initial version for the
new create file. And here the ci->i_inline_version will keep with 1,
it's buggy.
3, buffer write to the file immediately:
ceph_write_iter()-->
ceph_get_caps(file, need=Fw, want=Fb, ...);
generic_perform_write()-->
a_ops->write_begin()-->
ceph_write_begin()-->
netfs_write_begin()-->
netfs_begin_read()-->
netfs_rreq_submit_slice()-->
netfs_read_from_server()-->
rreq->netfs_ops->issue_read()-->
ceph_netfs_issue_read()-->
{
...
if (ci->i_inline_version != CEPH_INLINE_NONE &&
ceph_netfs_issue_op_inline(subreq))
return;
...
}
ceph_put_cap_refs(ci, Fwb);
The ceph_netfs_issue_op_inline() will send a getattr(Fsr) request to
mds.1.
4, then the mds.1 will request the rd lock for CInode::filelock from
the auth mds.0, the mds.0 will do the CInode::filelock state transation
from excl --> sync, but it need to revoke the Fxwb caps back from the
clients.
While the kernel client has aleady held the Fwb caps and waiting for
the getattr(Fsr).
It's deadlock!
URL: https://tracker.ceph.com/issues/55377
Signed-off-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
If any 'x' caps is issued we can just choose the auth MDS instead
of the random replica MDSes. Because only when the Locker is in
LOCK_EXEC state will the loner client could get the 'x' caps. And
if we send the getattr requests to any replica MDS it must auth pin
and tries to rdlock from the auth MDS, and then the auth MDS need
to do the Locker state transition to LOCK_SYNC. And after that the
lock state will change back.
This cost much when doing the Locker state transition and usually
will need to revoke caps from clients.
URL: https://tracker.ceph.com/issues/55240
Signed-off-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Pull page cache updates from Matthew Wilcox:
- Appoint myself page cache maintainer
- Fix how scsicam uses the page cache
- Use the memalloc_nofs_save() API to replace AOP_FLAG_NOFS
- Remove the AOP flags entirely
- Remove pagecache_write_begin() and pagecache_write_end()
- Documentation updates
- Convert several address_space operations to use folios:
- is_dirty_writeback
- readpage becomes read_folio
- releasepage becomes release_folio
- freepage becomes free_folio
- Change filler_t to require a struct file pointer be the first
argument like ->read_folio
* tag 'folio-5.19' of git://git.infradead.org/users/willy/pagecache: (107 commits)
nilfs2: Fix some kernel-doc comments
Appoint myself page cache maintainer
fs: Remove aops->freepage
secretmem: Convert to free_folio
nfs: Convert to free_folio
orangefs: Convert to free_folio
fs: Add free_folio address space operation
fs: Convert drop_buffers() to use a folio
fs: Change try_to_free_buffers() to take a folio
jbd2: Convert release_buffer_page() to use a folio
jbd2: Convert jbd2_journal_try_to_free_buffers to take a folio
reiserfs: Convert release_buffer_page() to use a folio
fs: Remove last vestiges of releasepage
ubifs: Convert to release_folio
reiserfs: Convert to release_folio
orangefs: Convert to release_folio
ocfs2: Convert to release_folio
nilfs2: Remove comment about releasepage
nfs: Convert to release_folio
jfs: Convert to release_folio
...
The pages in the file mapping maybe reclaimed and reused by other
subsystems and the page->private maybe used as flags field or
something else, if later that pages are used by page caches again
the page->private maybe not cleared as expected.
Here will check the PG_private bit instead of the folio->private.
Cc: stable@vger.kernel.org
URL: https://tracker.ceph.com/issues/55421
Signed-off-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Luis Henriques <lhenriques@suse.de>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Use a folio throughout ceph_release_folio().
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
With all implementations of aops->readpage converted to aops->read_folio,
we can stop checking whether it's set and remove the member from aops.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Change all the callers of ->readpage to call ->read_folio in preference,
if it exists. This is a transitional duplication, and will be removed
by the end of the series.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
There are no more aop flags left, so remove the parameter.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
There are no more aop flags left, so remove the parameter.
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Pull netfs updates from David Howells:
"Netfs prep for write helpers.
Having had a go at implementing write helpers and content encryption
support in netfslib, it seems that the netfs_read_{,sub}request
structs and the equivalent write request structs were almost the same
and so should be merged, thereby requiring only one set of
alloc/get/put functions and a common set of tracepoints.
Merging the structs also has the advantage that if a bounce buffer is
added to the request struct, a read operation can be performed to fill
the bounce buffer, the contents of the buffer can be modified and then
a write operation can be performed on it to send the data wherever it
needs to go using the same request structure all the way through. The
I/O handlers would then transparently perform any required crypto.
This should make it easier to perform RMW cycles if needed.
The potentially common functions and structs, however, by their names
all proclaim themselves to be associated with the read side of things.
The bulk of these changes alter this in the following ways:
- Rename struct netfs_read_{,sub}request to netfs_io_{,sub}request.
- Rename some enums, members and flags to make them more appropriate.
- Adjust some comments to match.
- Drop "read"/"rreq" from the names of common functions. For
instance, netfs_get_read_request() becomes netfs_get_request().
- The ->init_rreq() and ->issue_op() methods become ->init_request()
and ->issue_read(). I've kept the latter as a read-specific
function and in another branch added an ->issue_write() method.
The driver source is then reorganised into a number of files:
fs/netfs/buffered_read.c Create read reqs to the pagecache
fs/netfs/io.c Dispatchers for read and write reqs
fs/netfs/main.c Some general miscellaneous bits
fs/netfs/objects.c Alloc, get and put functions
fs/netfs/stats.c Optional procfs statistics.
and future development can be fitted into this scheme, e.g.:
fs/netfs/buffered_write.c Modify the pagecache
fs/netfs/buffered_flush.c Writeback from the pagecache
fs/netfs/direct_read.c DIO read support
fs/netfs/direct_write.c DIO write support
fs/netfs/unbuffered_write.c Write modifications directly back
Beyond the above changes, there are also some changes that affect how
things work:
- Make fscache_end_operation() generally available.
- In the netfs tracing header, generate enums from the symbol ->
string mapping tables rather than manually coding them.
- Add a struct for filesystems that uses netfslib to put into their
inode wrapper structs to hold extra state that netfslib is
interested in, such as the fscache cookie. This allows netfslib
functions to be set in filesystem operation tables and jumped to
directly without having to have a filesystem wrapper.
- Add a member to the struct added above to track the remote inode
length as that may differ if local modifications are buffered. We
may need to supply an appropriate EOF pointer when storing data (in
AFS for example).
- Pass extra information to netfs_alloc_request() so that the
->init_request() hook can access it and retain information to
indicate the origin of the operation.
- Make the ->init_request() hook return an error, thereby allowing a
filesystem that isn't allowed to cache an inode (ceph or cifs, for
example) to skip readahead.
- Switch to using refcount_t for subrequests and add tracepoints to
log refcount changes for the request and subrequest structs.
- Add a function to consolidate dispatching a read request. Similar
code is used in three places and another couple are likely to be
added in the future"
Link: https://lore.kernel.org/all/2639515.1648483225@warthog.procyon.org.uk/
* tag 'netfs-prep-20220318' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs:
afs: Maintain netfs_i_context::remote_i_size
netfs: Keep track of the actual remote file size
netfs: Split some core bits out into their own file
netfs: Split fs/netfs/read_helper.c
netfs: Rename read_helper.c to io.c
netfs: Prepare to split read_helper.c
netfs: Add a function to consolidate beginning a read
netfs: Add a netfs inode context
ceph: Make ceph_init_request() check caps on readahead
netfs: Change ->init_request() to return an error code
netfs: Refactor arguments for netfs_alloc_read_request
netfs: Adjust the netfs_failure tracepoint to indicate non-subreq lines
netfs: Trace refcounting on the netfs_io_subrequest struct
netfs: Trace refcounting on the netfs_io_request struct
netfs: Adjust the netfs_rreq tracepoint slightly
netfs: Split netfs_io_* object handling out
netfs: Finish off rename of netfs_read_request to netfs_io_request
netfs: Rename netfs_read_*request to netfs_io_*request
netfs: Generate enums from trace symbol mapping lists
fscache: export fscache_end_operation()
Pull ceph updates from Ilya Dryomov:
"The highlights are:
- several changes to how snap context and snap realms are tracked
(Xiubo Li). In particular, this should resolve a long-standing
issue of high kworker CPU usage and various stalls caused by
needless iteration over all inodes in the snap realm.
- async create fixes to address hangs in some edge cases (Jeff
Layton)
- support for getvxattr MDS op for querying server-side xattrs, such
as file/directory layouts and ephemeral pins (Milind Changire)
- average latency is now maintained for all metrics (Venky Shankar)
- some tweaks around handling inline data to make it fit better with
netfs helper library (David Howells)
Also a couple of memory leaks got plugged along with a few assorted
fixups. Last but not least, Xiubo has stepped up to serve as a CephFS
co-maintainer"
* tag 'ceph-for-5.18-rc1' of https://github.com/ceph/ceph-client: (27 commits)
ceph: fix memory leak in ceph_readdir when note_last_dentry returns error
ceph: uninitialized variable in debug output
ceph: use tracked average r/w/m latencies to display metrics in debugfs
ceph: include average/stdev r/w/m latency in mds metrics
ceph: track average r/w/m latency
ceph: use ktime_to_timespec64() rather than jiffies_to_timespec64()
ceph: assign the ci only when the inode isn't NULL
ceph: fix inode reference leakage in ceph_get_snapdir()
ceph: misc fix for code style and logs
ceph: allocate capsnap memory outside of ceph_queue_cap_snap()
ceph: do not release the global snaprealm until unmounting
ceph: remove incorrect and unused CEPH_INO_DOTDOT macro
MAINTAINERS: add Xiubo Li as cephfs co-maintainer
ceph: eliminate the recursion when rebuilding the snap context
ceph: do not update snapshot context when there is no new snapshot
ceph: zero the dir_entries memory when allocating it
ceph: move to a dedicated slabcache for ceph_cap_snap
ceph: add getvxattr op
libceph: drop else branches in prepare_read_data{,_cont}
ceph: fix comments mentioning i_mutex
...
Pull filesystem folio updates from Matthew Wilcox:
"Primarily this series converts some of the address_space operations to
take a folio instead of a page.
Notably:
- a_ops->is_partially_uptodate() takes a folio instead of a page and
changes the type of the 'from' and 'count' arguments to make it
obvious they're bytes.
- a_ops->invalidatepage() becomes ->invalidate_folio() and has a
similar type change.
- a_ops->launder_page() becomes ->launder_folio()
- a_ops->set_page_dirty() becomes ->dirty_folio() and adds the
address_space as an argument.
There are a couple of other misc changes up front that weren't worth
separating into their own pull request"
* tag 'folio-5.18b' of git://git.infradead.org/users/willy/pagecache: (53 commits)
fs: Remove aops ->set_page_dirty
fb_defio: Use noop_dirty_folio()
fs: Convert __set_page_dirty_no_writeback to noop_dirty_folio
fs: Convert __set_page_dirty_buffers to block_dirty_folio
nilfs: Convert nilfs_set_page_dirty() to nilfs_dirty_folio()
mm: Convert swap_set_page_dirty() to swap_dirty_folio()
ubifs: Convert ubifs_set_page_dirty to ubifs_dirty_folio
f2fs: Convert f2fs_set_node_page_dirty to f2fs_dirty_node_folio
f2fs: Convert f2fs_set_data_page_dirty to f2fs_dirty_data_folio
f2fs: Convert f2fs_set_meta_page_dirty to f2fs_dirty_meta_folio
afs: Convert afs_dir_set_page_dirty() to afs_dir_dirty_folio()
btrfs: Convert extent_range_redirty_for_io() to use folios
fs: Convert trivial uses of __set_page_dirty_nobuffers to filemap_dirty_folio
btrfs: Convert from set_page_dirty to dirty_folio
fscache: Convert fscache_set_page_dirty() to fscache_dirty_folio()
fs: Add aops->dirty_folio
fs: Remove aops->launder_page
orangefs: Convert launder_page to launder_folio
nfs: Convert from launder_page to launder_folio
fuse: Convert from launder_page to launder_folio
...
If read_mapping_folio() fails then "inline_version" is printed without
being initialized.
[ jlayton: use CEPH_INLINE_NONE instead of "-1" ]
Fixes: 083db6fd3e ("ceph: uninline the data on a file opened for writing")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
If a ceph file is made up of inline data, uninline that in the ceph_open()
rather than in ceph_page_mkwrite(), ceph_write_iter(), ceph_fallocate() or
ceph_write_begin().
This makes it easier to convert to using the netfs library for VM write
hooks.
Should this also take the inode lock for the duration on uninlining to
prevent a race with truncation?
[ jlayton: fix up folio locking, update i_inline_version after write ]
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
One fewer pointer dereference, and in the future we may not be able to
count on the mapping pointer being populated (e.g. in the DIO case).
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Pull ceph updates from Ilya Dryomov:
"One notable change here is that async creates and unlinks introduced
in 5.7 are now enabled by default. This should greatly speed up things
like rm, tar and rsync. To opt out, wsync mount option can be used.
Other than that we have a pile of bug fixes all across the filesystem
from Jeff, Xiubo and Kotresh and a metrics infrastructure rework from
Luis"
* tag 'ceph-for-5.16-rc1' of git://github.com/ceph/ceph-client:
ceph: add a new metric to keep track of remote object copies
libceph, ceph: move ceph_osdc_copy_from() into cephfs code
ceph: clean-up metrics data structures to reduce code duplication
ceph: split 'metric' debugfs file into several files
ceph: return the real size read when it hits EOF
ceph: properly handle statfs on multifs setups
ceph: shut down mount on bad mdsmap or fsmap decode
ceph: fix mdsmap decode when there are MDS's beyond max_mds
ceph: ignore the truncate when size won't change with Fx caps issued
ceph: don't rely on error_string to validate blocklisted session.
ceph: just use ci->i_version for fscache aux info
ceph: shut down access to inode when async create fails
ceph: refactor remove_session_caps_cb
ceph: fix auth cap handling logic in remove_session_caps_cb
ceph: drop private list from remove_session_caps_cb
ceph: don't use -ESTALE as special return code in try_get_cap_refs
ceph: print inode numbers instead of pointer values
ceph: enable async dirops by default
libceph: drop ->monmap and err initialization
ceph: convert to noop_direct_IO
Add proper error handling for when an async create fails. The inode
never existed, so any dirty caps or data are now toast. We already
d_drop the dentry in that case, but the now-stale inode may still be
around. We want to shut down access to these inodes, and ensure that
they can't harbor any more dirty data, which can cause problems at
umount time.
When this occurs, flag such inodes as being SHUTDOWN, and trash any caps
and cap flushes that may be in flight for them, and invalidate the
pagecache for the inode. Add a new helper that can check whether an
inode or an entire mount is now shut down, and call it instead of
accessing the mount_state directly in places where we test that now.
URL: https://tracker.ceph.com/issues/51279
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
We have our own op, but the WARN_ON is not terribly helpful, and it's
otherwise identical to the noop one. Just use that.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Pull ceph updates from Ilya Dryomov:
- a set of patches to address fsync stalls caused by depending on
periodic rather than triggered MDS journal flushes in some cases
(Xiubo Li)
- a fix for mtime effectively not getting updated in case of competing
writers (Jeff Layton)
- a couple of fixes for inode reference leaks and various WARNs after
"umount -f" (Xiubo Li)
- a new ceph.auth_mds extended attribute (Jeff Layton)
- a smattering of fixups and cleanups from Jeff, Xiubo and Colin.
* tag 'ceph-for-5.15-rc1' of git://github.com/ceph/ceph-client:
ceph: fix dereference of null pointer cf
ceph: drop the mdsc_get_session/put_session dout messages
ceph: lockdep annotations for try_nonblocking_invalidate
ceph: don't WARN if we're forcibly removing the session caps
ceph: don't WARN if we're force umounting
ceph: remove the capsnaps when removing caps
ceph: request Fw caps before updating the mtime in ceph_write_iter
ceph: reconnect to the export targets on new mdsmaps
ceph: print more information when we can't find snaprealm
ceph: add ceph_change_snap_realm() helper
ceph: remove redundant initializations from mdsc and session
ceph: cancel delayed work instead of flushing on mdsc teardown
ceph: add a new vxattr to return auth mds for an inode
ceph: remove some defunct forward declarations
ceph: flush the mdlog before waiting on unsafe reqs
ceph: flush mdlog before umounting
ceph: make iterate_sessions a global symbol
ceph: make ceph_create_session_msg a global symbol
ceph: fix comment about short copies in ceph_write_end
ceph: fix memory leak on decode error in ceph_handle_caps
Ceph has a following race between hole punching and page fault:
CPU1 CPU2
ceph_fallocate()
...
ceph_zero_pagecache_range()
ceph_filemap_fault()
faults in page in the range being
punched
ceph_zero_objects()
And now we have a page in punched range with invalid data. Fix the
problem by using mapping->invalidate_lock similarly to other
filesystems. Note that using invalidate_lock also fixes a similar race
wrt ->readpage().
CC: Jeff Layton <jlayton@kernel.org>
CC: ceph-devel@vger.kernel.org
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Jan Kara <jack@suse.cz>
This will collect IO's total size and then calculate the average
size, and also will collect the min/max IO sizes.
The debugfs will show the size metrics in bytes and will let the
userspace applications to switch to what they need.
URL: https://tracker.ceph.com/issues/49913
Signed-off-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
The sparse tool complains as follows:
fs/ceph/addr.c:316:37: warning:
symbol 'ceph_netfs_read_ops' was not declared. Should it be static?
This symbol is not used outside of addr.c, so mark it static.
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Reviewed-by: Ilya Dryomov <idryomov@gmail.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
The checks for page->mapping are odd, as set_page_dirty is an
address_space operation, and I don't see where it would be called on a
non-pagecache page.
The warning about the page lock also seems bogus. The comment over
set_page_dirty() says that it can be called without the page lock in
some rare cases. I don't think we want to warn if that's the case.
Reported-by: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Start preparing to allow the use of THPs in the pagecache with ceph by
making it use thp_size() in lieu of PAGE_SIZE in the appropriate places.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
All of the existing callers that don't set this to NULL just drop the
page reference at some arbitrary point later in processing. There's no
point in keeping a page reference that we don't use, so just drop the
reference immediately after checking the Uptodate flag.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
There is no need to do a ceph_pool_perm_check() on anything that isn't a
regular file, as the MDS is what handles talking to the OSD in those
cases. Just return 0 if it's not a regular file.
Reported-by: Luis Henriques <lhenriques@suse.de>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Convert ceph_readpages to ceph_readahead and make it use
netfs_readahead. With this we can rip out a lot of the old
readpage/readpages infrastructure.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Convert ceph_write_begin to use the netfs_write_begin helper. Most of
the ops we need for it are already in place from the readpage conversion
but we do add a new check_write_begin op since ceph needs to be able to
vet whether there is an incompatible writeback already in flight before
reading in the page.
With this, we can also remove the old ceph_do_readpage helper.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
Have the ceph KConfig select NETFS_SUPPORT. Add a new netfs ops
structure and the operations for it. Convert ceph_readpage to use
the new netfs_readpage helper.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>
